CN107094111B

CN107094111B - Label transfer method in software defined network controller deployment

Info

Publication number: CN107094111B
Application number: CN201710347355.1A
Authority: CN
Inventors: 汪清; 赵建军; 杨耀通; 高丽蓉; 刘文斌
Original assignee: Tianjin University
Current assignee: Tianjin University
Priority date: 2017-05-17
Filing date: 2017-05-17
Publication date: 2020-08-18
Anticipated expiration: 2037-05-17
Also published as: CN107094111A

Abstract

The invention belongs to the field of next generation network control and management and software defined network, and aims to ensure intra-domain connectivity after load balancing adjustment. The invention adopts the technical scheme that a label transmission method in software defined network controller deployment is characterized in that label transmission is carried out from a central point of each classified domain, the central point in the domain and other nodes have different labels, the labels of the central point are diffused to adjacent nodes one by one and replace the labels of the adjacent nodes until the labels of the whole network are not changed any more, after the operation, the nodes with the labels different from the central point are isolated nodes, the isolated nodes are distributed to classes/domains of the nodes different from the adjacent nodes, and the operation is repeated until no isolated node exists. The invention is mainly applied to the software defined network occasion.

Description

Label transfer method in software defined network controller deployment

Technical Field

The invention belongs to the field of next generation network control and management and software defined network, relates to a novel method for removing isolated nodes after SDN partitioning, and particularly relates to an algorithm for ensuring the mutual communication of nodes in each domain after network domain partitioning.

Background

Software Defined Networking (SDN) is emerging as a promising future network paradigm. A big difference between SDN and legacy networks is: the control plane becomes a set of dedicated controllers, each of which manages one or more simplified packet forwarding switches.

Although research in SDN has attracted much attention, many problems still exist. Taking the control plane as an example, given a network with a certain number of specific nodes, proper partitioning of the network is an important issue. To address this issue, many metrics such as latency, intra-domain connectivity, security, resource conservation, and load balancing are often considered. However, when the load balance of the entire network is ensured, the nodes in a certain domain are often split, so that the original intra-domain communication is changed into cross-domain communication, which causes a series of problems of performance, reliability, safety and the like. This is an intra-domain connectivity problem for the controller. This situation is easily encountered when the network is a non-convex structure. Therefore, intra-domain connectivity occupies a significant position in all problems.

Many scholars make a great deal of contribution to the problem of controller deployment in SDN. In document [1], Heller et al originally proposed the deployment problem of the controller, and analyzed the deployment problem of the controller using the average delay and the maximum delay as metrics, and solved it using a greedy algorithm. However, this document does not consider the load problem of the controller, and it is not applicable in large networks to deploy the controller only from a time delay point of view. Document [2] presents a complete model based on deployment costs in the controller deployment problem, but the document does not present an algorithm. Document [3] proposes a node calculation index of a pressure center and gives a controller deployment algorithm based on the center, but the document does not relate to the problems of controller capacity and balance. Document [4] proposes a K-critical algorithm, which calculates the number of required controllers and the deployment positions of the controllers according to the maximum allowable delay by constructing a Robust tree, but the document also does not consider the capacities of the controllers.

[1]HELLER B,SHERWOOD R,MCKEOWN N.The controller placement problem[C]//Proceedings of the First Workshop on Hot Topics in Software DefinedNetworks.ACM,2012:7-12

[2]SALLAHI A,ST-HILAIRE M.Optimal model for the controller placementproblem in software defined networks[J].Communications Letters,IEEE,2015,19(1):30-33

[3]ISHIGAKI G,SHINOMIYA N.Controller placement algorithm to alleviateburdens on communication nodes[C]//2016International Conference on Computing,Networking and Communications(ICNC).IEEE,2016:1-5

[4]

Y,

C,GARCIA AJ.On the controllerplacement for designing a distributed SDN control layer[C]//NetworkingConference,2014IFIP.IEEE,2014:1-9.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention aims to provide an isolated node algorithm after the domain division of a software defined network is removed, and the intra-domain connectivity after load balancing adjustment is ensured. The invention adopts the technical scheme that a label transmission method in software defined network controller deployment is characterized in that label transmission is carried out from a central point of each classified domain, the central point in the domain and other nodes have different labels, the labels of the central point are diffused to adjacent nodes one by one and replace the labels of the adjacent nodes until the labels of the whole network are not changed any more, after the operation, the nodes with the labels different from the central point are isolated nodes, the isolated nodes are distributed to classes/domains of the nodes different from the adjacent nodes, and the operation is repeated until no isolated node exists.

The tag transfer operation is as follows:

step 1: marking the central point as a black label, and marking the other nodes as white labels;

step 2: marking the node labels adjacent to the black nodes as black;

and step 3: repeating the step 2 until the node labels are not changed any more;

and 4, step 4: repeat

steps

1,2,3 for all domains.

In a specific example, a topological structure is set as G (V, E), V represents a node set, E represents an edge set, the total number of switches is N, K controllers are to be deployed, it is assumed that the whole network has been classified according to time delay and load balancing, at this time, the classification result is a center point set P and a label set Q of each node, each class/domain is called a subnet, and the number of an initial subnet is j ═ 1;

step 1: to subnet V_jPerforming label transfer operation with black label as central point, V_jThe rest nodes in the node list are white labels;

step 2: the number of black nodes L at this time is calculated₀Marking the labels of the white label nodes adjacent to the black label node into black;

and step 3: the number of black nodes L at this time is calculated₁If L is₀＝L₁All white nodes in the subnet at this moment are isolated nodes and are stored in a set I, and j is j + 1; otherwise, re-executing the

steps

1,2 and 3;

and 4, step 4: judging whether j is larger than K: if j is larger than K, returning to the isolated node set I; otherwise, repeating the

steps

1,2,3 and 4;

and 5: traversing all nodes in the set I, and distributing each point to the classes of adjacent nodes in different classes;

step 6: judging whether the set I is an empty set: if the current set is an empty set, returning the classification result at the moment, and ending the algorithm; otherwise, repeating the

steps

1,2,3,4 and 5.

The invention has the characteristics and beneficial effects that:

the invention has the advantages that after the network is adjusted according to the load balance, the isolated nodes are removed, and the intra-domain connectivity is ensured, so that the deployment result of the controller is more reasonable.

There have also been many studies by some scholars on the problems of SDN controller deployment. Among them, a lot of literature takes the time delay from the switch to the controller as the deployment basis. Some scholars use greedy algorithm, particle swarm algorithm, simulated annealing and other algorithms to solve the deployment problem of the controller. However, no relevant literature currently considers intra-domain connectivity. The algorithm makes full use of the thought of the breadth-first algorithm, and provides a label transfer algorithm to eliminate isolated nodes after the SDN controller is deployed. The process complexity is low, and the running time is extremely short.

In the aspect of solving the performance estimation of the result, the number of the isolated nodes is used as a measurement target in the algorithm. As a result, as shown in fig. 3 below, in comparison with an algorithm that does not consider intra-domain connectivity, taking the topology of the american Internet2 as an example, the number of isolated nodes is always 0 after the label passing algorithm is run.

Description of the drawings:

FIG. 1 is a flow chart of a staining algorithm.

Fig. 2 is a flow chart of a tag delivery algorithm.

FIG. 3 is a graph comparing the number of isolated nodes.

Detailed Description

The invention belongs to the field of next generation network control and management and software defined network, and relates to a novel algorithm for removing isolated nodes after SDN partitioning. And more particularly to an algorithm for ensuring that no orphaned nodes are present in each zone after network partitioning.

Although load balancing issues have been considered, no relevant literature has considered intra-domain connectivity issues after controller load balancing adjustments. After the load balancing is adjusted, nodes in a certain domain may be split, resulting in cross-domain communication. Therefore, the invention aims to provide an isolated node algorithm after removing the SDN partition, and ensure intra-domain connectivity after load balancing adjustment.

The main process of the invention is that label transmission is carried out from the central point of each classified sub-network, the central point in the domain and the rest other nodes are marked as labels with different colors, and the labels of the central point are transmitted to other nodes one by one until the labels of the whole network are not changed any more. After the above operation. And the node with the label different from the central point is the isolated node. These orphaned nodes are assigned to classes of nodes of different classes adjacent to them, and the operation is repeated until there are no orphaned nodes. The label transfer algorithm comprises the following specific steps:

step 1: the center point is marked as a black label, and the rest nodes are white labels.

Step 2: nodes adjacent to the black-labeled node are marked as black labels.

And step 3: repeat step 2 until all node labels are no longer changed.

And 4, step 4: repeat

steps

1,2,3 for all subnets.

The specific steps are shown in figure 1.

The specific scheme is as follows:

a label transfer algorithm in an SDN controller deployment:

assume a topology of G (V, E), where V represents a set of nodes and E represents a set of edges. The total number of the switches is N, and K controllers are deployed. Assume that the entire network has now been classified according to latency and load balancing, where the classification results are P (set of central points) and Q (set of labels per node).

Step 1: and performing label transfer operation on each subnet once, and storing all the isolated nodes in one set.

Step 2: these orphaned nodes are assigned to classes that neighbor nodes of different classes.

And step 3: judging whether the whole network has isolated nodes:

if no isolated node exists, returning to the adjusted classification; otherwise, returning to the step 1 to continue the algorithm.

The specific steps are shown in fig. 2.

The following is an example of the present invention.

A label transfer algorithm in an SDN controller deployment:

the initial classification is P (central point set) and Q (label set of each node), the total number of nodes is N, the whole network is divided into K classes, and the initial subnet number is j ═ 1.

Step 1: to subnet V_jA tag transfer operation is performed. The central point is a black label, V_jThe remaining nodes in (1) are white labels.

Step 2: the number of black nodes L at this time is calculated₀. The labels of the white label nodes adjacent to the black label node are marked in black.

And step 3: the number of black nodes L at this time is calculated₁. If L is₀＝L₁All white nodes in the subnet at this moment are isolated nodes and are stored in a set I, and j is j + 1; otherwise, the

steps

1,2 and 3 are executed again.

steps

1,2,3 and 4.

And 5: and traversing all nodes in the set I, and distributing each point into the classes of adjacent nodes in different classes.

steps

1,2,3,4 and 5.

Claims

1. A label transfer method in software defined network controller deployment is characterized in that label transfer is carried out from a central point of each classified domain, the central point in the domain and other nodes have different labels, the labels of the central point are diffused to adjacent nodes one by one to replace the labels of the adjacent nodes until the labels of the whole network are not changed any more, after the operation, the nodes with the labels different from the central point are isolated nodes, the isolated nodes are distributed to classes/domains of different classes of nodes adjacent to the isolated nodes, and the label transfer and the subsequent steps are repeated until no isolated nodes exist.

2. The method of claim 1, wherein the label delivery operation is as follows:

step 2: marking the node labels adjacent to the black nodes as black;

and 4, step 4: repeat steps 1,2,3 for all domains.

3. The method according to claim 1, wherein in a specific example, a topology is set as G (V, E), V denotes a node set, E denotes an edge set, the total number of switches is N, K controllers are to be deployed, and when the whole network is classified according to time delay and load balancing, the classification result is a central point set P and a label set Q of each node, each class/domain is referred to as a subnet, and an initial subnet number j is 1;

and step 3: the number of black nodes L at this time is calculated₁If L is₀＝L₁All white nodes in the subnet at this moment are isolated nodes and are stored in a set I, and j is j + 1; otherwise, re-executing the steps 1,2 and 3;

and 4, step 4: judging whether j is larger than K: if j is larger than K, returning to the isolated node set I; otherwise, repeating the steps 1,2,3 and 4;

step 6: judging whether the set I is an empty set: if the current set is an empty set, returning the classification result at the moment, and ending the algorithm; otherwise, repeating the steps 1,2,3,4 and 5.